mastodon 4.7.3Working on Theoretical Computer Science and Foundations of Machine Learning. Still also using twitter.
Excited about new work with Nikhil Vyas and Sham Kakade on formally defining and providing rigorous guarantees on potential copyright infringements by generative models. See the blog https://windowsontheory.org/2023/02/21/provable-copyright-protection-for-generative-models/
As we say in paper, there are many legal and ethical issues in generative models, and we are only focused on one of them. However I believe making a formal definition is important, also to ground future discussions.
Great Quanta article by @anil@sciencemastodon.com on diffusion-based generative models, giving a non-technical overview of the ingredients that came together to provide models such as Dall-E, Stable diffusion and Imagen https://www.quantamagazine.org/the-physics-principle-that-inspired-modern-ai-art-20230105/
I try to keep politics out of Mastodon but academic freedom is beyond politics. Whether it’s an art history professor fired for showing historical art at Hamline, or a human rights leader denied a visiting position at Harvard for advocating for human rights, too often university leaders opt for comfort over their core academic mission.
Scott Aaronson on the recent paper claiming factoring with sublinear number of qubits: "one of the most actively misleading quantum computing papers I’ve seen in 25 years, and I’ve seen … many"
A paperclip maximizer, a stochastic parrot, king Midas, a jpeg of the web, markets, democracy, an alien, a shoggoth,...
So many metaphors offered for AI. I discuss these and their limitations
https://windowsontheory.org/2023/06/28/metaphors-for-ai-and-why-i-dont-like-them/
Looking forward to talking in Sapienza University of Rome Tuesday June 20 on "The uneasy relation between deep learning and statistics"
https://di.uniroma1.it/it/notizie/seminari/distinguished-lectures
Once again, apologies about "politics" in Mastodon, but just wanted to support Black colleagues in the face of disgusting comments and quasi-apology from Bostrom.
New in FOCS 2023: A separate track for submitting papers on conjectures. https://windowsontheory.org/2023/01/16/new-in-focs-2023-a-conjectures-track/
@BostonAbrams@newsie.social @JamesGleick@zirk.us Actually ChatGPT is rather different than Trump. ChatGPT can compose pieces of text that on their own look plausible if you don't have any context. See my blog post https://windowsontheory.org/2023/01/03/memento-and-large-language-models/
Trump famously can't string two sentences together without contradicting himself.
First blog post of the new year, "Memento and Large Language Models" https://windowsontheory.org/2023/01/03/memento-and-large-language-models/
New paper on the implicit bias (or lack thereof) of stochastic gradient descent in the online learning setting https://arxiv.org/abs/2306.08590
We show that aside from the computational benefits, SGD noise (larger learning rate or smaller batches) do not offer implicit bias advantages.
With Nikhil Vyas, Depen Morwani, Rosie Zhao, Gal Kaplun, and Sham Kakade.
@BostonAbrams@newsie.social @JamesGleick@zirk.us There is some similarity, but I think Trump's statements attract people at an emotional level. I don't think I've seen a coherent paragraph written by him - he's at his best at slogans and tweets.
ChatGPT is more like the undergrad that doesn't know the material and fibs in the exam.
@matt_blackwell@mastodon.social I add the following to all my letters:
p.s. Due to the large number of reference letters I write, I cannot personally fill in application websites. Therefore, if your system uses such prompts, please ignore my responses. Hopefully, this letter contains enough information for you to be able to figure out the answers. In any case, feel free to email me with any questions or clarifications.
@abebab@scholar.social @gvdr@ecoevo.social This is my best friend in the morning https://us.moccamaster.com/products/cup-one